JP2000321176A - Error detection method and device - Google Patents

Abstract

(57) [Summary] [Problem] To determine whether an engine or the like is normal or abnormal without depending on human senses. An abnormality detection device includes a microphone.
Collects the engine sound of the engine 20 to extract the feature amount 1
Enter 4 The feature amount extraction unit 14 makes the feature amount obvious by spectrum analysis or the like, and inputs the feature amount to the type determination unit 26 (26a to 26n) of the type identification unit 16. Each type determiner 26 is provided corresponding to the type of the engine 20 and is formed of a neural network.
It is determined whether or not the engine 20 belongs to the corresponding type based on the input signal from. The abnormality detecting unit 18 checks the determination result of each type determiner 26 and determines which type determiner 26
The engine is determined to be an abnormal product when it is determined that the engine does not belong to the corresponding type, or when the two or more types determiner 26 determines that the engine belongs to the corresponding type.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abnormality detection method for detecting an abnormal article such as a defective article, and more particularly to an abnormality detection method suitable for detecting an abnormal article in a production line in which a plurality of types are mixed. The present invention relates to a detection method and device.

[0002]

2. Description of the Related Art Conventionally, in an automobile engine assembly factory, after assembling the engine, an inspector listens to the rotation of the engine and listens to the sound of the rotation. It is determined whether or not it has been done. The engine test includes a so-called cold test in which the engine is rotated by a motor and a so-called firing test in which the engine is actually operated by supplying fuel to the engine. The ring test is to be performed instead of the cold test.

[0003]

However, in the above-mentioned firing test, the working environment is not always good, for example, a loud noise is generated in the working room for actually operating the engine by burning the fuel. In addition, since the sensory test is based on the sound of the operator, a lot of experience is required to judge the quality of the engine, and the judgment may vary due to individual differences. There is also a worry that it will be connected. Further, in an automobile engine assembly factory, a plurality of types of engines corresponding to various types of automobiles often flow mixedly on the same line, and it is not easy to distinguish the engine sounds of these various types of engines. For this reason, there is a demand for the development of a method and a device capable of determining the quality of an engine without depending on human sensation.

On the other hand, conventionally, when judging the acceptability of a plurality of types of articles, generally, reference values (for example, length, weight, light intensity, voltage, etc.) determined in advance by an inspection device corresponding to each type are generally used. Is set and the measured value (detected value) actually obtained from the article is compared with the reference value to determine the quality. However, when a plurality of types are mixed, the number of installed inspection devices is determined. Increases the cost and increases the space. Further, in a production line where a plurality of types of articles flow in a mixed manner, when a single inspection device checks the quality of the products, a reference value for each type and a complex quality determination program based on the reference value are used. It is conceivable that the inspection is given to an inspection device.
However, in this case, when a new type is added, the program must be modified, which is troublesome and requires a lot of time to respond.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned drawbacks of the prior art, and has as its object to make it possible to determine whether an engine or the like is normal or abnormal without depending on human senses. Another object of the present invention is to make it possible to identify a plurality of types of articles when they are mixed, and to easily detect an abnormal product in the articles.

[0006]

In order to achieve the above object, an abnormality detection method according to the present invention acquires data capable of specifying a type from a plurality of types of inspection objects having different characteristics, and stores the data in the data. Based on which type it belongs to, and if it is determined that it does not belong to any type, or if it is determined that it belongs to two or more types,
Determining that the inspection object is abnormal;

[0007] An abnormality detecting apparatus for performing the above-described abnormality detecting method includes a detecting means for detecting a physical quantity capable of specifying a type from a plurality of types of inspection objects having different characteristics, and a detecting means corresponding to each of the types. And a type determination unit that receives the detection signal of the detection unit and processes the input detection signal by a neural network to determine whether the inspection object belongs to the type. When the judgment result of the type judging unit is inputted, and when it is judged from these judgment results that the object to be inspected does not belong to any of the above types, or when it is judged that it belongs to two or more types, the inspection object An abnormality detection unit that determines that the object is abnormal.

[0008]

According to the present invention constructed as described above, data capable of specifying the type of inspection object, for example, driving sound (engine sound) of an automobile engine, reflected light or transmitted light of pigment, dye, printing ink, etc., and furthermore, Light or the like generated when the alloy or the like is subjected to emission spectral analysis is detected, and the type of the inspection object is specified based on these data. Then, when the type of the inspection object is specified based on the above data, and it is determined that the inspection object does not belong to any type or belongs to two or more types, the inspection object is determined. The object is determined to be an abnormal item that is not normal.

That is, for example, when the object to be inspected is an engine such as an automobile, a cultivator, a generator, etc., the sound generated when the engine is driven depending on the size of the displacement, the size of the attached valve, the material, the structure of the engine, etc. Since the sound wave spectrum differs depending on the type of engine, the type can be specified by comparing the engine sound when the engine is driven with the sound of a normal engine that has been checked in advance. When the engine sound of the engine to be inspected does not belong to any type, or when it is determined that the engine belongs to two or more types, it is considered that a sound different from that of a normal engine is generated, and abnormal sound is generated. Can be regarded as goods.

[0010] This makes it possible to easily and surely detect abnormal products of the engine without depending on the sound of human beings, etc., thereby reducing the burden on the workers and eliminating inspection errors due to individual differences and fatigue. Can be. In addition, even if a plurality of types of articles are mixed and flowing, the inspection object can be identified for each type, and abnormal products such as defective products can be easily and reliably selected.

When identifying the type of the inspection object,
If performed in a type identification unit using a neural network, it is possible to reliably determine the type of an object, such as sound, light, or vibration spectrum, whose type cannot be specified by simply comparing it with a reference value (threshold). The type can be specified, and abnormal products can be reliably detected. Since the abnormality detection unit detects the abnormality of the inspection target based on the determination result of each type determination unit, even when a new type is added, it is necessary to rewrite a program for detecting the abnormality, No need to make any modifications,
I live by being able to respond easily.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of an abnormality detection method and apparatus according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 is a block diagram of an abnormality detection device according to an embodiment of the present invention, and shows an example applied to a case where an abnormal product of an automobile engine is detected. In FIG. 1, the abnormality detection device 10 includes a microphone 12 serving as a detection unit, a feature amount extraction unit 14, a type identification unit 16, and an abnormality detection unit 18 as main components. The microphone 12 is arranged in the vicinity of the engine 20 as an inspection object, collects sound (sound wave) emitted from the engine 20 as a physical quantity, converts the sound into an electric signal, and outputs a detection signal to an amplifier 22 connected to the output side. Is output.

The output side of the amplifier 22 has a normalizing circuit 24
Is provided. The normalization circuit 24 amplifies the entire input signal so that the maximum value of the signal input from the amplifier 22 becomes a constant value (voltage), and outputs the amplified signal to the feature amount extraction unit 14 connected to the output side. . The feature amount extraction unit 14 makes the feature amount obvious by analyzing signal components by, for example, spectrum analysis or sonagram analysis.

On the output side of the feature extraction unit 14, a type identification unit 16 is connected. The type identification unit 16 has a plurality of type determination units (type determination units) 26 (26a to 26n) corresponding to the types (classes) of the engine 20, and each of the type determination units 26 has a feature. The output signal of the quantity extractor 14 is input. Each type determiner 26 is composed of n types of, for example, hierarchical neural networks. These neural networks use a plurality of normal engines of a corresponding type to perform learning by engine sounds, and determine whether or not the corresponding type of engine is based on an input signal from the feature amount extraction unit 14. It can be determined. Also, the type determiners 26a-
An abnormality detection unit 18 is provided on the output side of 26n. The abnormality detection unit 18 determines whether the engine 20 is a normal product (non-defective product) or an abnormal product (defective product) based on an output signal of each type determination unit 26, as described in detail later.
The determination result is output to a display device, a printer, a storage device, and the like (not shown) as indicated by an arrow 28.

The operation of the embodiment configured as described above is as follows.
It is as follows. As shown in FIGS. 2A and 2B, the sound of the engine when the automobile engine is driven varies depending on the type of the engine (for example, displacement and structure), and sound waves of different frequencies are combined. As a result, an engine sound peculiar to the engine is obtained. Therefore, the engine sound when each type of normal engine is operated is made to be featured by spectrum analysis or sonagram analysis in advance, and given to each type determination unit 26 configured by a neural network to be learned. Thus, the characteristics of the engine sound for the corresponding type of engine are extracted and stored in each type determiner.

When determining whether the engine 20 is a normal product or an abnormal product, first, the engine 20 is operated to generate an engine sound, and the sound is collected by the microphone 12 and converted into an electric signal. . The electric signal (detection signal) output from the microphone 12 is amplified by the amplifier 22 and then supplied to the normalization circuit 24. The normalizing circuit 24 sets the maximum value of the signal input within a predetermined time (for example, one second) to a predetermined value (for example, “1”), and sets the maximum value in accordance with the size of the input signal for another signal. A signal having a magnitude corresponding to the ratio with respect to the value is sent to the feature amount extracting unit 14. The feature amount extraction unit 14 makes the feature amount of the engine sound obvious by spectrum analysis or sonogram analysis and outputs the result.

The output signal of the feature extracting unit 14 is input to each type determining unit 26 constituting the type identifying unit 16. Each type determiner 26 composed of a neural network determines whether or not the type belongs to the type learned by the engine 20 based on the feature amount revealed by the feature amount extraction unit 14,
As a result of the determination, for example, “1” is output to the abnormality detection unit 18 if it belongs, and “0” otherwise.

The abnormality detecting section 18 is provided with each type determining device 26a-26.
26n are sequentially examined, and if one of the type determiners 26 outputs a signal “1” belonging to the corresponding type and all the other type determiners 26 output “0”, It is determined (determined) that the engine 20 is a normal product. On the other hand, when all the type determination units 26 output “0” as not belonging to the corresponding type, the abnormality detection unit 18 determines that the engine is an abnormal product. further,
When two or more type determiners 26 output “1” as belonging to the corresponding types, the abnormality detection unit 18 does not belong to the two types at the same time because the engine 20 does not belong to the two types at the same time.
It is determined that the product is abnormal. These determination results are sent to a not-shown display device, printer, or the like.

As described above, in the embodiment, by learning and storing the characteristics of the engine sound corresponding to the type determiner 26 composed of the neural network provided for each type of the engine 20, the human sense of sound can be improved. The quality of the engine 20, that is, a normal product and an abnormal product can be distinguished without relying on a sensory test that relies on it, and an abnormal product can be easily and reliably detected. For this reason, it is possible to eliminate variations due to individual differences of workers and inspection errors due to work fatigue.
In addition, each type determination unit 26 determines only whether the type belongs to the corresponding type, and the abnormality detection unit 18 determines whether each type determination unit 26
Of the engine 20 based on the determination result of the above, even if a new type of engine is added, it is only necessary to provide a type determination unit corresponding to the new type of engine. It is possible to respond easily and promptly without having to do this. In the above embodiment, the maximum value of the output signal of the microphone 12 is set to a constant value by the normalization circuit 24. The influence on the determination result of the determiner 26 can be avoided.

In the above embodiment, the maximum value of the output signal of the microphone 12 is set to a constant value in the standardization circuit 24, but the standardization circuit 24 may be omitted. Further, in the above embodiment, the case where the inspection target is the engine 20 has been described, but the inspection target may be a machine or a plant. Further, in the above-described embodiment, the case where the physical quantity is the intensity of the sound wave has been described. However, for example, reflected light or transmitted light such as a dye or a pigment, or ink may be used, or light emitted from an alloy or the like may be used. The light may be emitted light for spectral analysis, or an abnormality may be detected by applying an impact to an automobile, a machine, a plant, or the like, and detecting these vibrations.

[0021]

As described above, according to the present invention, according to the present invention, data which can specify the type of the inspection object, such as the driving sound of an automobile engine (engine sound), or the color of pigment or dye, the printing ink, etc. The type of inspection object is specified based on the reflected light or transmitted light, or the light that was emitted when the alloy or the like was subjected to emission spectroscopic analysis, and the inspection object did not belong to any type. If it is determined that the inspection object does not exist or belongs to two or more types, the inspection object is determined to be an abnormal abnormality. Detection can be performed reliably, the burden on the operator can be reduced, and inspection errors due to individual differences and fatigue can be eliminated.
In addition, even if a plurality of types of articles are mixed and flowing, the inspection object can be identified for each type, and abnormal products such as defective products can be easily and reliably selected.

If the type of the object to be inspected is identified by the type identification unit using a neural network, the type can be identified by merely comparing it with a reference value (threshold), such as a spectrum of sound, light, or vibration. It is possible to reliably identify the type of goods that cannot be
An abnormal product can be reliably detected. Since the abnormality detection unit detects the abnormality of the inspection target based on the determination result of each type determination unit, even when a new type is added, it is necessary to rewrite a program for detecting the abnormality, There is no need to make corrections, and it is easy to respond and live.

[Brief description of the drawings]

FIG. 1 is a block diagram of an abnormality detection device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a difference in engine sound of an automobile engine.

[Explanation of symbols]

 Reference Signs List 10 abnormality detection device 12 detection means (microphone) 14 feature amount extraction unit 16 type identification unit 18 abnormality detection unit 20 inspection object (engine) 26a to 26n type determination unit

Claims (2)

[Claims] 1. A method for acquiring data capable of specifying a type from a plurality of types of inspection objects having different characteristics, determining which type the type belongs to based on the data, and determining that the type does not belong to any type. An abnormality detection method, wherein the inspection object is determined to be abnormal when the operation is performed or when the inspection object is determined to belong to two or more types. 2. A detection means for detecting a physical quantity capable of specifying a type from a plurality of types of inspection objects having different characteristics, and a detection signal provided for each of the types and receiving a detection signal of the detection means. A type determining unit that processes the input detection signal by a neural network to determine whether the inspection object belongs to the type, and a determination result of each type determining unit is input; When it is determined that the inspection target does not belong to any of the types, or when it is determined that the inspection target belongs to two or more types, an abnormality detection unit that determines that the inspection target is abnormal, An abnormality detection device comprising: